0.前言
     \quad\,\,\,\, 南开大学数学院投资学课程(本科生)笔记整理,不同于金融的投资学,此课程包含大量数学推导。第一部分包括关于投资组合的基础知识及现代投资组合理论(MPT) 的描述及模型推导。第二部分包括资本资产定价模型(CAPM)、套利定价理论(APT) 及资本市场线、证券市场线等概念及模型。
1.投资组合
-
资产
(1) Z = ∑ i = 1 n Z i Z = \sum_{i=1}^n Z_i \tag{1} Z=i=1∑nZi(1)
∗ Z *\quad Z ∗Z:总资产
∗ Z i *\quad Z_i ∗Zi:投资在证券i的资金
  \, -
投资策略集
(2) D = { x 1 , x 2 , ⋯   , x n   ∣   ∑ i = 1 n x i = 1 } D=\left\{x_1,x_2,\cdots,x_n\,\bigg|\,\sum_{i=1}^n x_i=1\right\} \tag{2} D={x1,x2,⋯,xn∣∣∣∣i=1∑nxi=1}(2)其中
(3) x i = Z i Z x_i=\frac{Z_i}{Z} \tag{3} xi=ZZi(3)
∗ x i < 0 *\quad x_i<0 ∗xi<0:允许卖空
∗ x i *\quad x_i ∗xi:比例,权重
证券1 | 证券2 | … \dots … | 证券k | … \dots … | 证券n | |
---|---|---|---|---|---|---|
收益率 | r 1 r_1 r1 | r 2 r_2 r2 | … \dots … | r k r_k rk | ⋯ \cdots ⋯ | r n r_n rn |
期望收益率 | E ( r 1 ) E(r_1) E(r1) | E ( r 2 ) E(r_2) E(r2) | ⋯ \cdots ⋯ | E ( r k ) E(r_k) E(rk) | … \dots … | E ( r n ) E(r_n) E(rn) |
标准差 | σ 1 \sigma_1 σ1 | σ 2 \sigma_2 σ2 | ⋯ \cdots ⋯ | σ k \sigma_k σk | … \dots … | σ n \sigma_n σn |
比例 | x 1 x_1 x1 | x 2 x_2 x2 | ⋯ \cdots ⋯ | x k x_k xk | … \dots … | x n x_n xn |
-
投资组合
  \, -
收益率
(4) r P = ∑ i = 1 n x i r i r_P=\sum_{i=1}^n x_ir_i \tag{4} rP=i=1∑nxiri(4) -
期望收益率
(5) E ( r P ) = ∑ i = 1 n x i ⋅ E ( r i ) E\left(r_P\right)=\sum_{i=1}^n x_i\cdot E\left(r_i\right) \tag{5} E(rP)=i=1∑nxi⋅E(ri)(5) -
风险
(6) σ P 2 = C o v ( ∑ i = 1 n x i r i ,   ∑ i = 1 n x i r i ) = ∑ i ∑ j x i x j σ i j = ∑ i x i 2 σ i 2 + 2 ∑ i ∑ j > i x i x j σ i σ j \begin{aligned} \sigma_P^2&=Cov\left(\sum_{i=1}^n x_ir_i,\,\sum_{i=1}^n x_ir_i\right)\\ &=\sum_{i}\sum_{j} x_ix_j\sigma_{ij}\\ &=\sum_{i} x_i^2\sigma_i^2+2\sum_{i}\sum_{j>i} x_ix_j\sigma_i\sigma_j \end{aligned} \tag{6} σP2=Cov(i=1∑nxiri,i=1∑nxiri)=i∑j∑xixjσij=i∑xi2σi2+2i∑j>i∑xixjσiσj(6)
σ i j = C o v ( r i ,   r j ) \quad\sigma_{ij}=Cov\left(r_i,\,r_j\right) σij=Cov(ri,rj)
σ i i = σ i 2 \quad\sigma_{ii}=\sigma_i^2 σii=σi2L e t Let Let
(7) X = ( x 1 , x 2 , ⋯   , x n ) T   r = ( r 1 , r 2 , ⋯   , r n ) T   R = ( E ( r 1 ) , E ( r 2 ) , ⋯   , E ( r n ) ) T   e = ( 1 , 1 , ⋯   , 1 ) T   V = ( σ i j ) n × n = [ σ 11 σ 12 ⋯ σ 1 n σ 21 σ 22 ⋯ σ 2 n ⋮ ⋮ ⋱ ⋮ σ n 1 σ n 2 ⋯ σ n n ]   n × n \begin{aligned} \textbf{X}&=\left(x_1,x_2,\cdots,x_n\right)^T \\\,\\ \textbf{r}&=\left(r_1,r_2,\cdots,r_n\right)^T\\\,\\ \textbf{R}&=\left(E\left(r_1\right),E\left(r_2\right),\cdots,E\left(r_n\right)\right)^T\\\,\\ \textbf{e}&=\left(1,1,\cdots,1\right)^T\\\,\\ \textbf{V}&=\left(\sigma_{ij}\right)_{n\times n}= \begin{bmatrix} \sigma_{11}&\sigma_{12}&\cdots&\sigma_{1n}\\ \sigma_{21}&\sigma_{22}&\cdots&\sigma_{2n}\\ \vdots&\vdots&\ddots&\vdots\\ \sigma_{n1}&\sigma_{n2}&\cdots&\sigma_{nn}\\ \end{bmatrix}_{\,n\times n} \end{aligned} \tag{7} XrReV=(x1,x2,⋯,xn)T=(r1,r2,⋯,rn)T=(E(r1),E(r2),⋯,E(rn))T=(1,1,⋯,1)T=(σij)n×n=⎣⎢⎢⎢⎡σ11σ21⋮σn1σ12σ22⋮σn2⋯⋯⋱⋯σ1nσ2n⋮σnn⎦⎥⎥⎥⎤n×n(7)
∗ V *\quad\textbf V ∗V为实对称正定矩阵H e n c e Hence Hence
(8) X T r = r P   X T R = E ( r P )   X T e = 1   X T VX = σ P 2 \begin{aligned}\textbf{X}^T\textbf{r}&=r_P\\\,\\\textbf{X}^T\textbf{R}&=E\left(r_P\right)\\\,\\ \textbf{X}^T\textbf{e}&=1\\\,\\\textbf{X}^T\textbf{V}\textbf{X}&=\sigma_P^2\end{aligned} \tag{8} XTrXTRXTeXTVX=rP=E(rP)=1=σP2(8)
-
分散化作用
当n充分大时, x i → 1 n x_i\rightarrow\frac{1}{n} xi→n1
于是有
(9) σ P 2 = ∑ i x i 2 σ i 2 + 2 ∑ i ∑ j > i x i x j σ i σ j → 1 n ( ∑ σ i 2 n ) + 2 ⋅ 1 n 2 ⋅ ∑ ∑ σ i j n 2 − n 2 ⋅ n 2 − n 2 \begin{aligned} \sigma_P^2 &=\sum_{i} x_i^2\sigma_i^2+2\sum_{i}\sum_{j>i} x_ix_j\sigma_i\sigma_j\\ &\rightarrow\frac{1}{n}\left(\frac{\sum\sigma_i^2}{n}\right)+2\cdot\frac{1}{n^2}\cdot\frac{\sum\sum\sigma_{ij}}{\frac{n^2-n}{2}}\cdot\frac{n^2-n}{2} \end{aligned} \tag{9} σP2=i∑xi2σi2+2i∑j>i∑xixjσiσj→n1(n∑σi2)+2⋅n21⋅2n2−n∑∑σij⋅2n2−n(9)我们称上式中的第一项为非系统性风险,第二项为系统性风险
可以看出,n充分大时,系统性风险逐渐趋于0,而非系统性风险趋于证券间的平均协方差
2.M-V模型
- 现代投资组合理论(Modern Portfolio Theory)
假设:
i.收益的度量
ii.风险的度量
iii.理性
iv.市场无摩擦
v.证券无限可分
-
投资组合选择原则风险偏好(风险偏好)
- 理性选择原则
- 不满足性:
投资者在其余条件相同的两个投资组合中进行选择时,总是选择预期回报率较高的组合。 - 厌恶风险:
投资者在其余条件相同的情况下,将选择标准差较小的组合。
- 不同的风险态度
厌恶风险
风险中性
爱好风险投资者的目标是投资效用最大化,而投资效用取决于预期收益率与风险。预期收益率带来正的效用,风险带来负的效用。引入无差异曲线以反映效用水平,一条无差异曲线代表给投资者带来同样满足程度的预期收益率和风险的所有组合。
-
无差异曲线
特征:
i.无差异曲线的斜率为正;
ii.无差异曲线是向下凸的;
iii.同一投资者有无限多条无差异曲线;
iv.同一投资者在同一时间、同一时点的任何两条无差异曲线都不能相交。∗ * ∗注意:无差异曲线的斜率越大,投资者越厌恶风险。
![](https://i-blog.csdnimg.cn/blog_migrate/b0b3eb9bb62811f127ff54e438b649ed.png)
-
投资效用函数( U \textbf{U} U)
(10) U = U ( R ˉ , σ ) U=U\left(\bar{R},\sigma\right) \tag{10} U=U(Rˉ,σ)(10)效用函数的形式多样,如
(11) U = R ˉ − 1 2 A σ 2 U=\bar{R}-\frac{1}{2}A\sigma^2 \tag{11} U=Rˉ−21Aσ2(11)其中, A A A表示投资者的风险厌恶系数,其典型值在 2 2 2至 4 4 4之间。
-
M-V模型描述
两证券组合P
(12) { r P = x 1 r 1 + x 2 r 2 E ( r P ) = x 1 E ( r 1 ) + x 2 E ( r 2 ) σ P 2 = x 1 2 σ 1 2 + x 2 2 σ 2 2 + 2 x 1 x 2 σ 12 1 = x 1 + x 2 \begin{cases} \begin{aligned} r_P&=x_1r_1+x_2r_2\\ E\left(r_P\right)&=x_1E\left(r_1\right)+x_2E\left(r_2\right)\\ \sigma_P^2&=x_1^2\sigma_1^2+x_2^2\sigma_2^2+2x_1x_2\sigma_{12}\\ 1&=x_1+x_2\\ \end{aligned} \end{cases} \tag{12} ⎩⎪⎪⎪⎪⎨⎪⎪⎪⎪⎧rPE(rP)σP21=x1r1+x2r2=x1E(r1)+x2E(r2)=x12σ12+x22σ22+2x1x2σ12=x1+x2(12)
消去 x 1 x_1 x1、 x 2 x_2 x2,得到以 σ P \sigma_P σP、 E ( r P ) E\left(r_P\right) E(rP)为横、纵坐标的双曲线的一支:对于多个证券,以 σ P \sigma_P σP、 E ( r P ) E\left(r_P\right) E(rP)为横、纵坐标,所有投资组合为双曲线一支围成的部分:
- 可行集的数学表达为:
(13)
G
=
{
(
σ
P
2
,
E
(
r
P
)
)
∣
{
E
(
r
P
)
=
X
T
R
σ
P
2
=
X
T
VX
,
X
∈
D
}
G=\left\{\left(\sigma_P^2,E\left(r_P\right)\right)\left|\left\{\begin{aligned}E\left(r_P\right)&=\textbf X^T\textbf R\\\sigma_P^2&=\textbf X^T\textbf V\textbf X\end{aligned}\right.\right.,\textbf X\in D\right\} \tag{13}
G={(σP2,E(rP))∣∣∣∣∣{E(rP)σP2=XTR=XTVX,X∈D}(13)
 
\,
- 其中投资组合的边界,成为最小方差组合(有效集):
  \,
(14) ∂ G = { ( σ P 2 , E ( r P ) ) ∣ m i n   σ P 2 = X T VX { X T R = μ X T e = 1 , μ 为 参 数 } \partial G=\left\{\left(\sigma_P^2,E\left(r_P\right)\right)\left|\begin{aligned}&{\rm min}\,\sigma_P^2=\textbf X^T\textbf V\textbf X\\&\begin{cases}\textbf X^T\textbf R=\mu\\\textbf X^T\textbf e=1\\\end{cases}\end{aligned}\right.,\mu为参数\right\} \tag{14} ∂G=⎩⎪⎪⎨⎪⎪⎧(σP2,E(rP))∣∣∣∣∣∣∣∣minσP2=XTVX{XTR=μXTe=1,μ为参数⎭⎪⎪⎬⎪⎪⎫(14)
![](https://i-blog.csdnimg.cn/blog_migrate/58ae6be658d1955dd740d5a529c24441.png)
-
全局最小方差组合   ( G l o b a l    M i n i m u m    V a r i a n c e    P o r t f o l i o )   \,(Global\,\,Minimum\,\,Variance\,\,Portfolio)\, (GlobalMinimumVariancePortfolio)为曲线左顶点,简记为 M V P MVP MVP。
-
位于 M V P MVP MVP上方的曲线称为有效前沿   ( E f f i c i e n t    F r o n t i e r )   \,(Efficient\,\,Frontier)\, (EfficientFrontier)
-
数学推导
Problem:
(15) m i n   σ P 2 = X T VX s . t .   { X T R = μ X T e = 1 \begin{aligned}&{\rm min}\,\sigma_P^2=\textbf X^T\textbf V\textbf X\\&s.t.\,\begin{cases}\textbf X^T\textbf R=\mu\\\textbf X^T\textbf e=1\\\end{cases}\end{aligned} \tag{15} minσP2=XTVXs.t.{XTR=μXTe=1(15)
Solution:L e t \quad Let Let
(16) L = X T VX − λ 1 ( X T R − μ ) − λ 2 ( X T e − 1 ) L=\textbf X^T\textbf V\textbf X-\lambda_1\left(\textbf X^T\textbf R-\mu\right)-\lambda_2\left(\textbf X^T\textbf e-1\right) \tag{16} L=XTVX−λ1(XTR−μ)−λ2(XTe−1)(16)H e n c e \quad Hence Hence
(17) ⟹ { ∂ L ∂ X = ( ∂ L ∂ X 1 , ⋯   , ∂ L ∂ X n ) T = 2 VX − λ 1 R − λ 2 e = 0 ∂ L ∂ λ 1 = − ( X T R − μ ) = 0 ∂ L ∂ λ 2 = − ( X T e − 1 ) = 0 \Longrightarrow\left\{\begin{aligned}&\frac{\partial L}{\partial\textbf X}=\left(\frac{\partial L}{\partial\textbf X_1},\cdots,\frac{\partial L}{\partial\textbf X_n}\right)^T=2\textbf V\textbf X-\lambda_1\textbf R-\lambda_2\textbf e=\textbf 0\\&\frac{\partial L}{\partial \lambda_1}=-\left(\textbf X^T\textbf R-\mu\right)=0\\&\frac{\partial L}{\partial \lambda_2}=-\left(\textbf X^T\textbf e-1\right)=0\end{aligned}\right. \tag{17} ⟹⎩⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎧∂X∂L=(∂X1∂L,⋯,∂Xn∂L)T=2VX−λ1R−λ2e=0∂λ1∂L=−(XTR−μ)=0∂λ2∂L=−(XTe−1)=0(17)T h u s \quad Thus Thus
(18) X = λ 1 2 V − 1 R + λ 2 2 V − 1 e \textbf X=\frac{\lambda_1}{2}\textbf V^{-1}\textbf R+\frac{\lambda_2}{2}\textbf V^{-1}\textbf e \tag{18} X=2λ1V−1R+2λ2V−1e(18)T h e n    w e    g e t \quad Then\,\,we\,\,get Thenweget
(19) { λ 1 2 R T V − 1 R + λ 2 2 e T V − 1 R = μ λ 1 2 R T V − 1 e    +   λ 2 2 e T V − 1 e = 1 \left\{\begin{aligned} \frac{\lambda_1}{2}\textbf R^T\textbf V^{-1}\textbf R+\frac{\lambda_2}{2}\textbf e^T\textbf V^{-1}\textbf R&=\mu\\ \frac{\lambda_1}{2}\textbf R^T\textbf V^{-1}\textbf e\,\,+\,\frac{\lambda_2}{2}\textbf e^T\textbf V^{-1}\textbf e&=1\\ \end{aligned}\right. \tag{19} ⎩⎪⎪⎨⎪⎪⎧2λ1RTV−1R+2λ2eTV−1R2λ1RTV−1e+2λ2eTV−1e=μ=1(19)L e t \quad Let Let
(20) { A = R T V − 1 R B = e T V − 1 R = ( e T V − 1 R ) T = R T V − 1 e C = e T V − 1 e D = A C − B 2 \left\{\begin{aligned} A&=\textbf R^T\textbf V^{-1}\textbf R\\ B&=\textbf e^T\textbf V^{-1}\textbf R=\left(\textbf e^T\textbf V^{-1}\textbf R\right)^T=\textbf R^T\textbf V^{-1}\textbf e\\ C&=\textbf e^T\textbf V^{-1}\textbf e\\ D&=AC-B^2 \end{aligned}\right. \tag{20} ⎩⎪⎪⎪⎪⎨⎪⎪⎪⎪⎧ABCD=RTV−1R=eTV−1R=(eTV−1R)T=RTV−1e=eTV−1e=AC−B2(20)
* ( V − 1 ) T = V − 1 \quad\quad\textbf *\quad\left(\textbf V^{-1}\right)^T=\textbf V^{-1} *(V−1)T=V−1(21) ( 19 ) ⟹ { A λ 1 + B λ 2 = 2 μ B λ 1 + C λ 2 = 2 \quad(19)\Longrightarrow \left\{\begin{aligned} A\lambda_1+B\lambda_2&=2\mu\\ B\lambda_1+C\lambda_2&=2\\ \end{aligned}\right. \tag{21} (19)⟹{Aλ1+Bλ2Bλ1+Cλ2=2μ=2(21)
  \,
I t ′ s    e a s y    t o    f i n d    t h a t    m a t r i x \quad It's\,\,easy\,\,to\,\,find\,\,that\,\,matrix It′seasytofindthatmatrix
(22) ( A B B C ) \begin{pmatrix}A&B\\B&C\end{pmatrix} \tag{22} (ABBC)(22)
i s    p o s i t i v e    d e f i n i t e .   B e c a u s e \quad is\,\,positive\,\,definite.\, Because ispositivedefinite.Because
(23) ( x y ) ( A B B C ) ( x y ) = ( x R + y e ) T V − 1 ( x R + y e ) \begin{pmatrix}x&y\end{pmatrix}\begin{pmatrix}A&B\\B&C\end{pmatrix}\begin{pmatrix}x\\y\end{pmatrix}=\left(x\textbf R+y\textbf e\right)^T\textbf V^{-1}\left(x\textbf R+y\textbf e\right) \tag{23} (xy)(ABBC)(xy)=(xR+ye)TV−1(xR+ye)(23)
S o    t h e    l i n e a r    e q u a t i o n    ( 21 )    h a s    t h e    o n l y    s o l u t i o n \quad So\,\,the\,\,linear\,\,equation\,\,(21)\,\,has\,\,the\,\,only\,\,solution Sothelinearequation(21)hastheonlysolution
(24) { λ 1 = 2 μ C − 2 B D λ 2 = 2 μ B − 2 A D \left\{\begin{aligned}\lambda_1&=\frac{2\mu C-2B}{D}\\\lambda_2&=\frac{2\mu B-2A}{D}\end{aligned}\right. \tag{24} ⎩⎪⎪⎨⎪⎪⎧λ1λ2=D2μC−2B=D2μB−2A(24)
T h u s \quad Thus Thus
(25) X ∗ = μ C − B D V − 1 R − μ B − A D V − 1 e \textbf X^*=\frac{\mu C-B}{D}\textbf V^{-1}\textbf R-\frac{\mu B-A}{D}\textbf V^{-1}\textbf e \tag{25} X∗=DμC−BV−1R−DμB−AV−1e(25)(26) m i n   σ P 2 = ( X ∗ ) T V ( X ∗ ) = ( X ∗ ) T [ μ C − B D R − μ B − A D e ] = μ C − B D μ − μ B − A D \begin{aligned}\rm{min}\,\sigma_P^2&=\left(\textbf X^*\right)^T\textbf V\left(\textbf X^*\right)\\&=\left(\textbf X^*\right)^T\left[\frac{\mu C-B}{D}\textbf R-\frac{\mu B-A}{D}\textbf e\right]\\&=\frac{\mu C-B}{D}\mu-\frac{\mu B-A}{D}\end{aligned} \tag{26} minσP2=(X∗)TV(X∗)=(X∗)T[DμC−BR−DμB−Ae]=DμC−Bμ−DμB−A(26)
(27) E ( r P ) = ( X ∗ ) T R = μ E\left(r_P\right)=\left(\textbf X^*\right)^T\textbf R=\mu \tag{27} E(rP)=(X∗)TR=μ(27)
  \,
a n d \quad and and
(28) σ P 2 = E ( r P ) C − B D E ( r P ) − E ( r P ) B − A D \sigma_P^2=\frac{E\left(r_P\right) C-B}{D}E\left(r_P\right)-\frac{E\left(r_P\right) B-A}{D} \tag{28} σP2=DE(rP)C−BE(rP)−DE(rP)B−A(28)  \,
W e    h a v e \quad We\,\,have Wehave
(29) σ P 2 1 C − ( E ( r P ) − B C ) 2 D C 2 = 1 \dfrac{\sigma_P^2}{\frac{1}{C}}-\dfrac{\left(E\left(r_P\right)-\frac{B}{C}\right)^2}{\frac{D}{C^2}}=1 \tag{29} C1σP2−C2D(E(rP)−CB)2=1(29)S o    w e    c a n    c o n c l u d e    t h e    c o o r d i n a t e s    o f    M V P    a r e \quad So\,\,we\,\,can\,\,conclude\,\,the\,\,coordinates\,\,of\,\,MVP\,\,are SowecanconcludethecoordinatesofMVPare
(30) ( 1 C ,   B C ) \left(\frac{1}{\sqrt C},\,\frac{B}{C}\right)\tag{30} (C1,CB)(30) -
事实上,我们也可以利用二次规划对   M V P   \,MVP\, MVP的坐标进行求解:
Problem:
(31) m i n σ P 2 = X T V − 1 X s . t .   X T e = 1 \begin{aligned}&{\rm min}\sigma_P^2=\textbf X^T\textbf V^{-1}\textbf X\\&s.t.\,\textbf X^T\textbf e=1\end{aligned}\tag{31} minσP2=XTV−1Xs.t.XTe=1(31)Solution:
L e t \quad Let Let
(32) L = X T VX − λ ( X T e − 1 ) L=\textbf X^T\textbf V\textbf X-\lambda\left(\textbf X^T\textbf e-1\right)\tag{32} L=XTVX−λ(XTe−1)(32)
H e n c e \quad Hence Hence
(33) ⟹ { ∂ L ∂ X = ( ∂ L ∂ X 1 , ⋯   , ∂ L ∂ X n ) T = 2 VX − λ e = 0 ∂ L ∂ λ = − ( X T e − 1 ) = 0 \Longrightarrow\left\{\begin{aligned}\frac{\partial L}{\partial\textbf X}&=\left(\frac{\partial L}{\partial\textbf X_1},\cdots,\frac{\partial L}{\partial\textbf X_n}\right)^T=2\textbf V\textbf X-\lambda\textbf e=\textbf 0\\\frac{\partial L}{\partial \lambda}&=-\left(\textbf X^T\textbf e-1\right)=0\end{aligned}\right.\tag{33} ⟹⎩⎪⎪⎨⎪⎪⎧∂X∂L∂λ∂L=(∂X1∂L,⋯,∂Xn∂L)T=2VX−λe=0=−(XTe−1)=0(33)
T h u s \quad Thus Thus
(34) X = λ 2 V − 1 e \textbf X=\frac{\lambda}{2}\textbf V^{-1}\textbf e\tag{34} X=2λV−1e(34)
I t    f o l l o w s    f r o m    ( 33 )    t h a t \quad It\,\,follows\,\,from\,\,\left(33\right)\,\,that Itfollowsfrom(33)that
X T e = λ 2 e T V − 1 e = λ 2 C = 1 \textbf X^T\textbf e=\frac{\lambda}{2}\textbf e^T\textbf V^{-1}\textbf e=\frac{\lambda}{2}C=1 XTe=2λeTV−1e=2λC=1
T h u s \quad Thus Thus
(35) X ∗ = λ 2 V − 1 e = V − 1 e C \textbf X^*=\frac{\lambda}{2}\textbf V^{-1}\textbf e=\frac{\textbf V^{-1}\textbf e}{C}\tag{35} X∗=2λV−1e=CV−1e(35)(36) m i n   σ P 2 = ( X ∗ ) T V ( X ) = 1 C ( X ∗ ) T e = 1 C \begin{aligned}{\rm min}\,\sigma_P^2&=\left(\textbf X^*\right)^T\textbf V\left(\textbf X\right)\\&=\frac 1C\left(\textbf X^*\right)^T\textbf e=\frac 1C\end{aligned}\tag{36} minσP2=(X∗)TV(X)=C1(X∗)Te=C1(36)
(37) E ( r P ) = ( X ∗ ) T R = e T V − 1 R C = B C E\left(r_P\right)=\left(\textbf X^*\right)^T\textbf R=\frac{\textbf e^T\textbf V^{-1}\textbf R}{C}=\frac BC\tag{37} E(rP)=(X∗)TR=CeTV−1R=CB(37)
S o    w e    c a n    c o n c l u d e    t h e    c o o r d i n a t e s    o f    M V P    a r e \quad So\,\,we\,\,can\,\,conclude\,\,the\,\,coordinates\,\,of\,\,MVP\,\,are SowecanconcludethecoordinatesofMVPare
( 1 C ,   B C ) \left(\frac{1}{\sqrt C},\,\frac{B}{C}\right) (C1,CB) -
存在无风险证券的情况
无风险证券收益率为 r f r_f rf,满足:
σ f 2 = 0 \sigma_f^2=0 σf2=0E ( r f ) = r f E\left(r_f\right)=r_f E(rf)=rf
记无风险证券投资比例为 x 0 x_0 x0,此时二次规划问题转化为:
Problem:
(38) m i n   σ P 2 = X T VX s . t .   { x 0 r f + X T R = μ x 0 + X T e = 1 \begin{aligned}&{\rm min}\,\sigma_P^2=\textbf X^T\textbf V\textbf X\\&s.t.\,\begin{cases}x_0r_f+\textbf X^T\textbf R=\mu\\x_0+\textbf X^T\textbf e=1\\\end{cases}\end{aligned} \tag{38} minσP2=XTVXs.t.{x0rf+XTR=μx0+XTe=1(38)Solution:
(39) ( 38 )   ⟹   X T ( R − e r f ) = ( μ − r f ) \left(38\right)\,\Longrightarrow\,\textbf X^T\left(\textbf R-\textbf er_f\right)=\left(\mu-r_f\right)\tag{39} (38)⟹XT(R−erf)=(μ−rf)(39)
L e t \quad Let Let
(40) L = X T VX − λ [ X T ( R − e r f ) − ( μ − r f ) ] L=\textbf X^T\textbf V\textbf X-\lambda\left[\textbf X^T\left(\textbf R-\textbf er_f\right)-\left(\mu-r_f\right)\right] \tag{40} L=XTVX−λ[XT(R−erf)−(μ−rf)](40)H e n c e \quad Hence Hence
(41) ⟹ { ∂ L ∂ X = ( ∂ L ∂ X 1 , ⋯   , ∂ L ∂ X n ) T = 2 VX − λ ( R − e r f ) = 0 ∂ L ∂ λ = − [ X T ( R − e r f ) − ( μ − r f ) ] = 0 \Longrightarrow\left\{\begin{aligned}\frac{\partial L}{\partial\textbf X}&=\left(\frac{\partial L}{\partial\textbf X_1},\cdots,\frac{\partial L}{\partial\textbf X_n}\right)^T=2\textbf V\textbf X-\lambda\left(\textbf R-\textbf er_f\right)=\textbf 0\\\frac{\partial L}{\partial \lambda}&=-\left[\textbf X^T\left(\textbf R-\textbf er_f\right)-\left(\mu-r_f\right)\right]=0\end{aligned}\right.\tag{41} ⟹⎩⎪⎪⎨⎪⎪⎧∂X∂L∂λ∂L=(∂X1∂L,⋯,∂Xn∂L)T=2VX−λ(R−erf)=0=−[XT(R−erf)−(μ−rf)]=0(41)T h u s \quad Thus Thus
(42) X = λ 2 V − 1 ( R − e r f ) \textbf X=\frac{\lambda}{2}\textbf V^{-1}\left(\textbf R-\textbf er_f\right)\tag{42} X=2λV−1(R−erf)(42)I t    f o l l o w s    f r o m    ( 39 )    t h a t \quad It\,\,follows\,\,from\,\,\left(39\right)\,\,that Itfollowsfrom(39)that
(43) X T ( R − e r f ) = λ 2 ( R − e r f ) T V − 1 ( R − e r f ) = ( μ − r f ) \textbf X^T\left(\textbf R-\textbf er_f\right)=\frac{\lambda}{2}\left(\textbf R-\textbf er_f\right)^T\textbf V^{-1}\left(\textbf R-\textbf er_f\right)=\left(\mu-r_f\right)\tag{43} XT(R−erf)=2λ(R−erf)TV−1(R−erf)=(μ−rf)(43)L e t \quad Let Let
(44) H = ( R − e r f ) T V − 1 ( R − e r f ) > 0 H=\left(\textbf R-\textbf er_f\right)^T\textbf V^{-1}\left(\textbf R-\textbf er_f\right)>0\tag{44} H=(R−erf)TV−1(R−erf)>0(44)T h u s \quad Thus Thus
(45) X ∗ = λ 2 V − 1 ( R − e r f ) = μ − r f H V − 1 ( R − e r f ) \begin{aligned}\textbf X^*&=\frac{\lambda}{2}\textbf V^{-1}\left(\textbf R-\textbf er_f\right)\\&=\frac{\mu-r_f}H\textbf V^{-1}\left(\textbf R-\textbf er_f\right)\end{aligned}\tag{45} X∗=2λV−1(R−erf)=Hμ−rfV−1(R−erf)(45)(46) m i n   σ P 2 = ( X ∗ ) T V ( X ∗ ) = μ − r f H ( X ∗ ) T ( R − e r f ) = ( μ − r f ) 2 H \begin{aligned}\rm{min}\,\sigma_P^2&=\left(\textbf X^*\right)^T\textbf V\left(\textbf X^*\right)\\&=\frac{\mu-r_f}H\left(\textbf X^*\right)^T\left(\textbf R-\textbf er_f\right)\\&=\frac{\left(\mu-r_f\right)^2}H\end{aligned} \tag{46} minσP2=(X∗)TV(X∗)=Hμ−rf(X∗)T(R−erf)=H(μ−rf)2(46)
(47) E ( r P ) = μ E\left(r_P\right)=\mu \tag{47} E(rP)=μ(47)
  \,
W e    h a v e \quad We\,\,have Wehave
(48) σ P = ± E ( r P ) − r f H \sigma_P=\pm\frac{E\left(r_P\right)-r_f}{\sqrt H}\tag{48} σP=±HE(rP)−rf(48)可以看出,再加入无风险证券后,最小方差组合由双曲线的一支变为一条折线,其顶点坐标位为 y y y轴上的点 ( 0 ,   r f ) \left(0,\,r_f\right) (0,rf),位于顶点上方的射线为有效前沿。
  \, -
两个最小方差组合的联系
\quad
由   ( 29 )   \,\left(29\right)\, (29)及   ( 48 )   \,\left(48\right)\, (48)可知:(49) σ P 2 1 C − ( E ( r P ) − B C ) 2 D C 2 = [ E ( r P ) − r f ] 2 H 1 C − ( E ( r P ) − B C ) 2 D C 2 = 1 \dfrac{\sigma_P^2}{\frac{1}{C}}-\dfrac{\left(E\left(r_P\right)-\frac{B}{C}\right)^2}{\frac{D}{C^2}}=\frac{\frac{\left[E\left(r_P\right)-r_f\right]^2}H}{\frac{1}{C}}-\dfrac{\left(E\left(r_P\right)-\frac{B}{C}\right)^2}{\frac{D}{C^2}}=1 \tag{49} C1σP2−C2D(E(rP)−CB)2=C1H[E(rP)−rf]2−C2D(E(rP)−CB)2=1(49)
其中,由   ( 44 ) \,\left(44\right) (44):
(50) H = ( R − e r f ) T V − 1 ( R − e r f ) = R T V − 1 R − r f e T V − 1 R − r f R T V − 1 e + r f 2 e T V − 1 e = A − r f B − r f B + r f 2 C = A + r f 2 C − 2 r f B \begin{aligned}H&=\left(\textbf R-\textbf er_f\right)^T\textbf V^{-1}\left(\textbf R-\textbf er_f\right)\\&=\textbf R^T\textbf V^{-1}\textbf R-r_f\textbf e^T\textbf V^{-1}\textbf R-r_f\textbf R^T\textbf V^{-1}\textbf e+r_f^2\textbf e^T\textbf V^{-1}\textbf e\\&=A-r_fB-r_fB+r_f^2C\\&=A+r_f^2C-2r_fB\end{aligned}\tag{50} H=(R−erf)TV−1(R−erf)=RTV−1R−rfeTV−1R−rfRTV−1e+rf2eTV−1e=A−rfB−rfB+rf2C=A+rf2C−2rfB(50)及   ( 20 ) \,\left(20\right) (20):
D = A C − B 2 D=AC-B^2 D=AC−B2
代入   ( 49 ) \,\left(49\right) (49)得:
( E ( r P ) − r f ) 2 A C + r f 2 − 2 r f B C − ( E ( r P ) − B C ) 2 A C − B 2 C 2 = 1 \frac{\left(E\left(r_P\right)-r_f\right)^2}{\frac AC+r_f^2-2r_f\frac BC}-\dfrac{\left(E\left(r_P\right)-\frac{B}{C}\right)^2}{\frac AC-\frac{B^2}{C^2}}=1 CA+rf2−2rfCB(E(rP)−rf)2−CA−C2B2(E(rP)−CB)2=1
整理得:
(51) [ ( E ( r P ) − B C ) ( r f − B C ) + D C 2 ] 2 = 0 \left[\left(E\left(r_P\right)-\frac BC\right)\left(r_f-\frac BC\right)+\frac D{C^2}\right]^2=0\tag{51} [(E(rP)−CB)(rf−CB)+C2D]2=0(51)a )   r f = B C   a)\,r_f=\frac BC\, a)rf=CB时,方程   ( 51 )   \,\left(51\right)\, (51)无解,说明此时两条曲线不交,这时容易验证曲线   ( 48 )   \,\left(48\right)\, (48)恰为曲线   ( 29 )   \,\left(29\right)\, (29)的两条渐近线。
  \,
b )   r f ≠ B C   b)\,r_f\ne\frac BC\, b)rf̸=CB时,方程   ( 51 )   \,\left(51\right)\, (51)有重根:
(52) E ( r P ) = B C + D C 2 B C − r f = A − B r f B − C r f \begin{aligned}E\left(r_P\right)&=\frac BC+\frac{\frac D{C^2}}{\frac BC-r_f}\\&=\frac{A-Br_f}{B-Cr_f}\end{aligned}\tag{52} E(rP)=CB+CB−rfC2D=B−CrfA−Brf(52)
\quad 说明此时两条曲线相切。
  \,
r f < B C   \quad r_f<\frac BC\, rf<CB时,曲线   ( 48 )   \,\left(48\right)\, (48)的上半部分与曲线   ( 29 )   \,\left(29\right)\, (29)的上半部分相切,如图:
r f > B C   \quad r_f>\frac BC\, rf>CB时,曲线   ( 48 )   \,\left(48\right)\, (48)的下半部分与曲线   ( 29 )   \,\left(29\right)\, (29)的下半部分相切。实际情况下,无风险收益   r f   \,r_f\, rf一般低于   M V P   \,MVP\, MVP的收益,我们只需要考虑上述情况中   r f < B C   \,r_f<\frac BC\, rf<CB的情况即可。
我们称切线为资本市场线   ( C a p i t a l    M a r k e t    L i n e   , C M L )   \,\left(Capital\,\,Market\,\,Line\,,CML\right)\, (CapitalMarketLine,CML),切点   P   \,P\, P为最优风险资产组合   ( O p t i m a l    R i s k y    P o r t f o l i o ) . \,\left(Optimal\,\,Risky\,\,Portfolio\right). (OptimalRiskyPortfolio).
-
事实上,在已知其相切的条件下,我们还可以利用二次规划对切点   P   \,P\, P的坐标进行求解:
Problem:
(53) m i n E ( r P ) − r f σ P s . t .   X T e = 1 {\rm min}\frac{E\left(r_P\right)-r_f}{\sigma_P}\\s.t.\,\textbf X^T\textbf e=1\tag{53} minσPE(rP)−rfs.t.XTe=1(53)
f o r \quad for for
E ( r P ) = X T R E\left(r_P\right)=\textbf X^T\textbf R E(rP)=XTR
a n d \quad and andσ P = X T VX \sigma_P=\sqrt{\textbf X^T\textbf{V}\textbf{X}} σP=XTVX
Solution:
L e t \quad Let Let
(54) L = E ( r P ) − r f σ P − λ ( X T e − 1 ) L=\frac{E\left(r_P\right)-r_f}{\sigma_P}-\lambda\left(\textbf X^T\textbf e-1\right)\tag{54} L=σPE(rP)−rf−λ(XTe−1)(54)H e n c e \quad Hence Hence
⟹   { ∂ L ∂ X = ( ∂ L ∂ X 1 , ⋯   , ∂ L ∂ X n ) T = ∂ ∂ X E ( r P ) ⋅ σ P − ( E ( r P ) − r f ) ⋅ ∂ ∂ X σ P σ P 2 − λ e = R X T VX − ( E ( r P ) − r f ) VX X T VX σ P 2 − λ e = σ P   R − E ( r P ) − r f σ P   VX σ P 2 − λ e = 0 ∂ L ∂ λ = − ( X T e − 1 ) = 0 \Longrightarrow\,\begin{cases}\begin{aligned}\frac{\partial L}{\partial\textbf X}&=\left(\frac{\partial L}{\partial X_1},\cdots,\frac{\partial L}{\partial X_n}\right)^T\\&=\frac{\frac{\partial}{\partial\textbf X}E\left(r_P\right)\cdot\sigma_P-\left(E\left(r_P\right)-r_f\right)\cdot\frac{\partial}{\partial\textbf X}\sigma_P}{\sigma_P^2}-\lambda\textbf e\\&=\frac{\textbf R\sqrt{\textbf X^T\textbf V\textbf X}-\left(E\left(r_P\right)-r_f\right)\frac{\textbf V\textbf X}{\sqrt{\textbf X^T\textbf V\textbf X}}}{\sigma_P^2}-\lambda \textbf e\\&=\frac{\sigma_P\,\textbf R-\frac {E\left(r_P\right)-r_f}{\sigma_P}\,\textbf V\textbf X}{\sigma_P^2}-\lambda \textbf e\\&=0\\\frac{\partial L}{\partial \lambda}&=-\left(\textbf X^T\textbf e-1\right)=0\end{aligned}\end{cases} ⟹⎩⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎧∂X∂L∂λ∂L=(∂X1∂L,⋯,∂Xn∂L)T=σP2∂X∂E(rP)⋅σP−(E(rP)−rf)⋅∂X∂σP−λe=σP2RXTVX−(E(rP)−rf)XTVXVX−λe=σP2σPR−σPE(rP)−rfVX−λe=0=−(XTe−1)=0
T h u s \quad Thus Thus
(55) ⟹   { σ P 2   R − ( E ( r P ) − r f ) VX − λ σ P 3   e = 0 X T e − 1 = 0 \Longrightarrow\,\begin{cases}\sigma_P^2\,\textbf R-\left(E\left(r_P\right)-r_f\right)\textbf {VX}-\lambda\sigma_P^3\,\textbf e=0\\\textbf X^T\textbf e-1=0\end{cases}\tag{55} ⟹{σP2R−(E(rP)−rf)VX−λσP3e=0XTe−1=0(55)a ) \quad a) a)
⟹ X T [ σ P 2   R − ( E ( r P ) − r f ) VX − λ σ P 3   e ] = σ P 2   X T R − ( E ( r P ) − r f ) X T VX − λ σ P 3   X T e = σ P 2   E ( r P ) − ( E ( r P ) − r f ) σ P 2 − λ σ P 3 = 0 \begin{aligned}\quad\Longrightarrow&\textbf X^T\left[\sigma_P^2\,\textbf R-\left(E\left(r_P\right)-r_f\right)\textbf {VX}-\lambda\sigma_P^3\,\textbf e\right]\\=&\sigma_P^2\,\textbf X^T\textbf R-\left(E\left(r_P\right)-r_f\right)\textbf X^T\textbf {VX}-\lambda\sigma_P^3\,\textbf X^T\textbf e\\=&\sigma_P^2\,E\left(r_P\right)-\left(E\left(r_P\right)-r_f\right)\sigma_P^2-\lambda\sigma_P^3\\=&0\end{aligned} ⟹===XT[σP2R−(E(rP)−rf)VX−λσP3e]σP2XTR−(E(rP)−rf)XTVX−λσP3XTeσP2E(rP)−(E(rP)−rf)σP2−λσP30
t h a t    i s \quad\quad that\,\,is thatis
(56) r f = λ   σ P r_f=\lambda\,\sigma_P\tag{56} rf=λσP(56)b ) \quad b) b)
⟹ R T V − 1 [ σ P 2   R − ( E ( r P ) − r f ) VX − λ σ P 3   e ] = σ P 2   R T V − 1 R − ( E ( r P ) − r f ) R T X − λ σ P 3   R T V − 1 e = A   σ P 2 − ( E ( r P ) − r f ) E ( r P ) − λ   B σ P 3 = 0 \begin{aligned}\quad\Longrightarrow&\textbf R^T\textbf V^{-1}\left[\sigma_P^2\,\textbf R-\left(E\left(r_P\right)-r_f\right)\textbf {VX}-\lambda\sigma_P^3\,\textbf e\right]\\=&\sigma_P^2\,\textbf R^T\textbf V^{-1}\textbf R-\left(E\left(r_P\right)-r_f\right)\textbf R^T\textbf X-\lambda \sigma_P^3\,\textbf R^T \textbf V^{-1}\textbf e\\=&A\,\sigma_P^2-\left(E\left(r_P\right)-r_f\right)E\left(r_P\right)-\lambda\,B\sigma_P^3\\=&0\end{aligned} ⟹===RTV−1[σP2R−(E(rP)−rf)VX−λσP3e]σP2RTV−1R−(E(rP)−rf)RTX−λσP3RTV−1eAσP2−(E(rP)−rf)E(rP)−λBσP30
t h a t    i s \quad\quad that\,\,is thatis
(57) ( E ( r P ) − r f ) E ( r P ) = σ P 2 ( A − λ   B σ P ) \left(E\left(r_P\right)-r_f\right)E\left(r_P\right)=\sigma_P^2\left(A-\lambda\,B\sigma_P\right)\tag{57} (E(rP)−rf)E(rP)=σP2(A−λBσP)(57)
c ) \quad c) c)
⟹ e T V − 1 [ σ P 2   R − ( E ( r P ) − r f ) VX − λ σ P 3   e ] = σ P 2   e T V − 1 R − ( E ( r P ) − r f ) e T X − λ σ P 3   e T V − 1 e = B   σ P 2 − ( E ( r P ) − r f ) − λ   C σ P 3 = 0 \begin{aligned}\Longrightarrow&\textbf e^T\textbf V^{-1}\left[\sigma_P^2\,\textbf R-\left(E\left(r_P\right)-r_f\right)\textbf{VX}-\lambda\sigma_P^3\,\textbf e\right]\\=&\sigma_P^2\,\textbf e^T\textbf V^{-1}\textbf R-\left(E\left(r_P\right)-r_f\right)\textbf e^T\textbf X-\lambda\sigma_P^3\,\textbf e^T\textbf V^{-1}\textbf e\\=&B\,\sigma_P^2-\left(E\left(r_P\right)-r_f\right)-\lambda\,C\sigma_P^3\\=&0\end{aligned} ⟹===eTV−1[σP2R−(E(rP)−rf)VX−λσP3e]σP2eTV−1R−(E(rP)−rf)eTX−λσP3eTV−1eBσP2−(E(rP)−rf)−λCσP30
t h a t    i s \quad\quad that\,\,is thatis
(58) E ( r P ) − r f = σ P 2 ( B − λ   C σ P ) E\left(r_P\right)-r_f=\sigma_P^2\left(B-\lambda\,C\sigma_P\right)\tag{58} E(rP)−rf=σP2(B−λCσP)(58)
F r o m   ( 56 ) ∼ ( 58 )   w e    k n o w \quad From\,(56)\sim(58)\,we\,\,know From(56)∼(58)weknow
E ( r P ) A − λ   B σ P = σ P 2 E ( r P ) − r f = 1 B − λ   C σ P \frac{E\left(r_P\right)}{A-\lambda\,B\sigma_P}=\frac{\sigma_P^2}{E\left(r_P\right)-r_f}=\frac{1}{B-\lambda\,C\sigma_P} A−λBσPE(rP)=E(rP)−rfσP2=B−λCσP1(59) ⟹   E ( r P ) = A − B ( λ σ P ) B − C ( λ σ P ) = A − B r f B − C r f \begin{aligned}\Longrightarrow\,E\left(r_P\right)&=\frac{A-B\left(\lambda\sigma_P\right)}{B-C\left(\lambda\sigma_P\right)}\\&=\frac{A-Br_f}{B-Cr_f}\end{aligned}\tag{59} ⟹E(rP)=B−C(λσP)A−B(λσP)=B−CrfA−Brf(59)
I t    f o l l o w s    f r o m    ( 50 ) ,   ( 58 )    &    ( 59 )    t h a t It\,\,follows\,\,from\,\,\left(50\right),\,\left(58\right)\,\,\&\,\,\left(59\right)\,\,that Itfollowsfrom(50),(58)&(59)that
(60) σ P 2 = E ( r P ) − r f B − λ   C σ P = E ( r P ) − r f B − C r f = A − B r f B − C r f − r f B − C r f = A + r f 2 C − 2 r f B ( B − C r f ) 2 = H ( B − C r f ) 2 \begin{aligned}\sigma_P^2&=\frac{E\left(r_P\right)-r_f}{B-\lambda\,C\sigma_P}=\frac {E\left(r_P\right)-r_f}{B-Cr_f}\\&=\frac{\frac{A-Br_f}{B-Cr_f}-r_f}{B-Cr_f}=\frac{A+r_f^2C-2r_fB}{\left(B-Cr_f\right)^2}\\&=\frac{H}{\left(B-Cr_f\right)^2}\end{aligned}\tag{60} σP2=B−λCσPE(rP)−rf=B−CrfE(rP)−rf=B−CrfB−CrfA−Brf−rf=(B−Crf)2A+rf2C−2rfB=(B−Crf)2H(60)
S o    w e    c a n    c o n c l u d e    t h e    c o o r d i n a t e s    o f    P    a r e So\,\,we\,\,can\,\,conclude\,\,the\,\,coordinates\,\,of\,\,P\,\,are SowecanconcludethecoordinatesofPare
(61) ( A − B r f B − C r f ,   H B − C r f ) \left(\frac{A-Br_f}{B-Cr_f},\,\frac{\sqrt{H}}{B-Cr_f}\right)\tag{61} (B−CrfA−Brf,B−CrfH)(61)